Humanoid robot hand designers are making one of their most consequential hardware choices: keep the motors away from the fingers and transmit force through tendons, or place actuators closer to the joints they move. Neither architecture has won, because each moves the engineering difficulty somewhere else.

Comparison of tendon-driven and in-finger actuator architectures for humanoid robot hands — Two broad approaches to packaging actuation in humanoid hands. The comparison is useful, but product architectures and revisions evolve; verify each manufacturer's current transmission design before treating a model as a fixed example.

A robot hand has to produce useful grip force while remaining light, responsive, compact, durable and safe around people. It also needs position feedback, often tactile sensing, manageable heat, practical wiring and a service plan. Those goals pull the design in different directions.

The central trade-off: tendon-driven hands move mass and heat away from the fingers but add transmission and calibration complexity. In-finger or in-hand actuation shortens the path between actuator and joint but puts mass, heat and hardware into the most space-constrained part of the robot.

Approach 1: Tendon-driven robot hands

In a tendon-driven hand, motors or linear actuators are mounted in the forearm, wrist or palm. Cables route their force to the finger joints in much the same way that human muscles pull tendons. 1X publicly describes NEO's hand as tendon-driven, with the motors in the forearm and proprietary tendons passing through the wrist.

Where tendon drive is strongest

Lower distal mass: removing motors and gearboxes from the fingers reduces inertia. The hand can change direction faster, and a collision carries less moving energy.
Human-like proportions: narrow fingers leave more room for tactile sensors, compliant pads and joint structure.
Remote heat management: motors can share a larger forearm structure, heat spreader or cooling path instead of heating the fingers.
Mechanical compliance: tendon elasticity can soften contact and help a finger conform to irregular objects, provided the compliance is modeled and controlled.
Actuator freedom: the drive unit is not forced to fit inside a phalange. Designers can choose a motor, rotary spool or feedback linear actuator based on force and stroke.

What tendon drive makes harder

Calibration and hysteresis: cable stretch, sheath compression, pulley friction and routing changes mean a given motor movement does not always produce exactly the same joint movement in both directions.
Pretension: tendons need the correct tension. Too little creates slack and poor response; too much increases friction, bearing load and energy consumption.
Routing: every bend consumes space and can add friction. Routing through a moving wrist is particularly demanding.
Antagonistic motion: a tendon primarily pulls. Returning a joint usually requires an opposing tendon, spring or another mechanical return path.
Service labor: motors may be centralized, but replacing or retensioning a cable can require opening the wrist and tracing a tightly packed path through the hand.
Control-model complexity: the controller must estimate the relationship between actuator position, tendon tension and joint angle as the hand changes posture and load.

The result is a hand that can be light and dexterous at the fingers, but only if the transmission is manufactured, routed and calibrated very well. The complexity has not disappeared; it has moved into the tendons, guides, tensioning system and control software.

Approach 2: In-finger or in-hand actuators

The alternative places each motor or actuator in the palm, at the finger base or inside the phalange it moves. The transmission can be a compact gearbox, screw, linkage or direct-drive joint. ROBOTIS documents the HX5-D20 as using a direct-drive actuator at each powered finger joint. Unitree's Dex5-1 specifications describe hollow-cup motors and independently replaceable fingers, illustrating the modular direction available to integrated designs.

Where integrated actuation is strongest

More direct kinematics: actuator position maps more directly to joint position, reducing the uncertainty caused by long cable paths.
Modular assembly: a finger can be built and tested as a subassembly. A damaged module may be swapped without rerouting tendons through the forearm.
Simpler commissioning: local joint limits, encoders and drive electronics can make calibration more repeatable from one hand to another.
Independent joints: local actuation can reduce mechanical coupling and make it easier to command one joint without disturbing another.
Shorter force path: fewer pulleys and cable bends can reduce transmission friction and tension-management hardware.

What integrated actuation makes harder

Finger mass and inertia: motors, gears, screws and housings move with the finger. That can reduce acceleration, increase impact energy and require more torque upstream at the wrist and arm.
Thermal concentration: heat is generated inside a small, enclosed mechanism that may be touching an object or a person.
Packaging: actuators compete with bearings, wiring, tactile sensors and structural material. Human-scale proportions leave little spare volume.
Bulk and dexterity: larger joints or thicker fingers can limit the ability to reach into tight spaces or wrap around small objects.
Distributed electronics: more local actuators mean more wires, connectors, motor drivers and failure points inside a moving hand.
Shock exposure: joint actuators sit closer to impacts and grip loads. Gear teeth, lead screws and bearings must survive those loads without excessive backlash.

This architecture simplifies the relationship between the motor and the joint, but it makes the mechanical and thermal packaging problem much less forgiving.

Trade-off comparison

Design factor	Tendon-driven	In-finger or in-hand
Finger mass	Usually lower because the primary actuator is remote.	Usually higher because drive hardware moves with the finger.
Dexterity and proportions	Supports slimmer, lower-inertia fingers, but routing can constrain wrist motion.	Direct joint control can be precise, but packaging may thicken fingers and joints.
Control	Must compensate for stretch, friction, slack, coupling and hysteresis.	More direct position mapping, but many local joints still require coordinated control.
Heat	Can be moved into the forearm or palm where cooling area is larger.	Concentrated in the hand, where cooling space is limited.
Efficiency	Loses energy through bends, guides, tendon deformation and pretension.	Avoids long cable losses but still has gearbox, bearing or screw friction.
Compliance	Can provide useful passive compliance, but that same compliance reduces positioning certainty.	Can feel mechanically rigid unless compliance is designed into the joint or controller.
Maintenance	Centralized actuators are accessible, while tendon replacement and retensioning can be labor-intensive.	Modules may be easy to swap, but the hand contains more motors, connectors and precision parts.
Typical failure modes	Frayed or stretched tendons, tension drift, pulley wear and routing friction.	Gear wear, backlash, overheated motors, damaged wiring and impact-loaded bearings.
Manufacturing	Demands repeatable cable lengths, routing and calibration.	Demands highly compact precision modules and tight thermal integration.
Best fit	Hands prioritizing low distal mass, slim fingers and compliant dexterity.	Hands prioritizing modular joints, direct control and fast service replacement.

The two categories are becoming a spectrum

Real hands do not have to live at either extreme. A hybrid design can keep high-power actuators in the forearm, place smaller actuators in the palm, and use passive or underactuated joints near the fingertips. It can also combine tendons for flexion with springs for extension, or use local actuators only where independent motion creates the most value.

That is why model names alone can be misleading. The supplied comparison graphic is a useful way to visualize the debate, but public specifications do not disclose every transmission detail and designs change between generations. For example, 1X explicitly documents forearm-mounted tendon motors, while AGIBOT publishes a motor-and-screw linkage specification for OmniHand Pro. Figure's published material emphasizes tactile fingertips and palm sensing, and Sharpa emphasizes active degrees of freedom and tactile sensing, without fully documenting every actuator location. The engineering principles are more durable than any one product label.

Where a feedback micro linear actuator fits

A compact feedback linear actuator can serve either architecture because it converts electrical power into a controlled straight-line pull or push. In a tendon-driven hand, it can sit in the forearm or palm and pull a tendon, move a cable tensioner or drive a compact linkage. In an integrated design, it can drive a palm linkage or a larger finger module directly.

The FIRGELLI Micro Pen Linear Actuator with feedback has a 16 mm body diameter, 20-100 mm stroke options and force options from 4.5 to 22 lb. Its dual Hall-effect feedback provides position pulses for closed-loop control, with published repeatability of 0.1 mm. Those characteristics make it useful when the design needs compact linear motion and measured actuator position.

Important packaging reality: depending on stroke, the Micro Pen's retracted length is approximately 140-220 mm. In a human-scale hand, that generally favors forearm, wrist or palm installation rather than placement inside a small distal finger segment. A larger robotic gripper or finger module may provide more room.

Feedback tells the controller where the actuator is; it does not by itself measure fingertip force or object contact. A dexterous hand will normally combine actuator-position feedback with current monitoring, tendon-tension sensing, joint sensing or tactile sensors. The actuator's 10% duty-cycle rating also needs to be included in the motion profile and thermal validation. Fast, continuous finger cycling may require a different actuator technology or a design that spreads the work across multiple mechanisms.

For tendon operation, remember that the actuator's linear stroke must be translated into the required tendon excursion, and a return spring or antagonistic tendon is usually needed. For direct linkage operation, calculate the changing mechanical advantage throughout the entire joint travel; peak force often occurs away from the visually obvious end position.

How to choose an architecture

Start with the task. A service robot handling towels and cups has different priorities from an industrial hand repeatedly gripping a fixture.
Budget distal mass. Count everything that moves with each finger, including motors, housings, fasteners, wiring and sensors.
Model the complete force path. Include pulley friction, tendon stretch, linkage leverage, joint losses and worst-case grip geometry.
Plan for heat. Compare peak motion with continuous use, and identify where heat can leave the mechanism.
Define the sensing stack. Position, current, tension and tactile feedback answer different questions; one does not replace the others.
Design the repair procedure early. Decide whether technicians will replace a tendon, a finger module or the entire hand, and how recalibration will work afterward.
Test wear and drift. Cycle the hand under realistic load and contamination while tracking backlash, tendon tension, temperature and positioning error.

So which approach wins?

Neither, at least not universally. Tendon-driven hands are compelling when low finger inertia, slim proportions and compliant dexterity dominate the requirements. Integrated actuators are compelling when direct control, modular assembly and field service matter more than minimum distal mass. Hybrids may become the most practical answer because they let designers put each actuator where its mass, heat and service requirements are easiest to manage.

The open question in 2026 is not simply whether a robot hand should use tendons or in-finger motors. It is where the designer should place mass, heat, compliance, sensing and maintenance complexity for the work that hand must perform. Compact feedback actuators widen the available design space, but the winning hand will still be the one whose complete mechanical, electrical and control system has been engineered as a whole.

Explore compact feedback actuation

Review force, stroke, speed, dimensions, Hall-feedback specifications and duty-cycle limits for the FIRGELLI Micro Pen.

View Micro Pen Actuators

Sources and further reading

Share This Article

Tendon-Driven vs In-Finger Actuators in Humanoid Robot Hands